Skip to content

Run a virtual iPhone on an Apple Silicon Mac.

Swift

Fill gaps in the template, clone and service notes

Jacky1cb62d1e95e9
diff --git a/Research/Guest/service_trimming.md b/Research/Guest/service_trimming.md
index 1d08c74d..708a0a39 100644
--- a/Research/Guest/service_trimming.md
+++ b/Research/Guest/service_trimming.md
@@ -306,3 +306,23 @@ frames, because the iPhone 27.0 guest publishes no synthetic camera
(`libvcamcaptured` logs `_sSourceList not located` on every boot, before and
after), so no client can reach the drive; and the microphone, whose code did
not change.
+
+## Open items
+
+- **iPadOS.** vphoned picks the list by major version only, so iPad guests on
+ 26 and 27 get the iPhone lists. No iPad guest has been measured with them.
+- **WidgetKit control extensions on iOS 26.** The `ControlAction.init(_:)`
+ crashes on 2 of 16 trimmed 26.6.2 boots ([iOS 26](#ios-26)) point at a
+ disabled job an App Intent reference needs; which one was not found.
+- **Location on iOS 27.** Location simulation under the profile is verified
+ on 26.6.2 only; on 27.0 it waits for locationd itself to work
+ ([locationd startup deadlock](locationd_startup_deadlock.md)).
+- **Host memory.** The VM has no balloon device, so the guest's freed memory
+ stays with the VM process (8.67 → 7.26 GB on 27.0). Returning it would take
+ `VZVirtioTraditionalMemoryBalloonDevice` and a guest that drives it, which
+ was not evaluated; giving trimmed machines less memory is the simple way to
+ use the headroom.
+- **Sample size.** Each configuration's idle figures come from one or two
+ 15-minute windows, and bursts of maintenance work that `dasd` schedules
+ (mediaanalysisd, cloudd, searchd) dominated some of them. They show the
+ direction, not a precise rate.
diff --git a/Research/Guest/template_snapshot_deletion.md b/Research/Guest/template_snapshot_deletion.md
index d82cd38d..102c3732 100644
--- a/Research/Guest/template_snapshot_deletion.md
+++ b/Research/Guest/template_snapshot_deletion.md
@@ -121,8 +121,26 @@ still writing about 2 GB while the snapshot went. The day before, the same call
from a vphoned with only `com.apple.developer.vfs.snapshot` answered
`not_permitted`.
-## Not yet verified
-
-- The reclaim of a deletion made by vphoned inside a template build has not
- been read off the host image; the 1.19 GB above comes from the prototype run
- after an offline trim.
+## Reclaim in a template build
+
+Measured on 2026-10-08 in a template build (iPhone17,3 iOS 27.0 24A435,
+standard trim, then the setup boot with vphoned deleting the snapshot). The
+host image cannot isolate it, because the setup boot writes first-boot state
+at the same time, so the System volume's Capacity Consumed was read from the
+stopped image (`diskutil apfs list` on a read-only attach) at each stage:
+
+| Stage | System volume | Data | User |
+| --- | --- | --- | --- |
+| After `cfw install` and Launchpad's first boot | 16,209,444,864 B | 1.17 GB | 0.46 GB |
+| After `vm template trim --tier standard` (1.166 GB of files) | 16,209,846,272 B | | |
+| After the setup boot | 15,022,374,912 B | 1.67 GB | 0.69 GB |
+
+The System volume shrank by 1,187,471,360 B (1.19 GB): the 1.166 GB trimmed
+plus about 21.6 MB held only by the snapshot, the same split as the prototype
+run. On the host, `Disk.img` (sampled every 2 s) went from 17.948 GB to
+16.620 GB within about 10 s of the snapshot step, then grew back with the setup
+boot's writes to 17.545 GB at shutdown. The 1.33 GB drop is 0.14 GB more
+than the System volume's, probably discards from the other volumes in the
+same seconds. The frozen template was 17.58 GB, against 18.92 GB for a
+template with neither trim nor setup boot and 19.3 GB for one with a setup
+boot and no trim.
diff --git a/Research/Guest/virtual_camera_transport.md b/Research/Guest/virtual_camera_transport.md
index d27c2fdb..327467d7 100644
--- a/Research/Guest/virtual_camera_transport.md
+++ b/Research/Guest/virtual_camera_transport.md
@@ -138,3 +138,8 @@ the life of their process:
After both changes `cameracaptured` took about one wakeup a second in the same
window, and AccessibilityUIServer and InputUI dropped from about 900 each to
under 10. The numbers are in [`service_trimming.md`](service_trimming.md#idle-cost-of-vphones-own-components).
+Camera frames were not checked live after the change: an iPhone 27.0 guest
+publishes no synthetic camera at all (`libvcamcaptured` logs `_sSourceList not
+located` on every boot), so nothing can start the drive there. That is a
+separate open item ([iOS 27 capture sources](ios27_capture_microphone_source.md#4-the-virtual-camera-does-not-change-this)).
+On 26.6.2 the synthetic source is published and the Camera app starts.
diff --git a/Research/Host/machine_identity_and_clone.md b/Research/Host/machine_identity_and_clone.md
index d30051e3..a38511a8 100644
--- a/Research/Host/machine_identity_and_clone.md
+++ b/Research/Host/machine_identity_and_clone.md
@@ -127,8 +127,10 @@ and a clone's disk image:
`nvram.bin` and `SEPStorage` are cloned too and diverge by a few blocks.
A clone therefore costs about 0.35 GB after its first boots and grows with
use, where a restore writes about 20 GB plus an 11 GB temporary restore tree.
-`du` and Launchpad's size column report the full 20 GB for every clone, and
-deleting the source frees no shared blocks while a clone remains.
+`du` reports the full 20 GB for every clone (Launchpad's Exclusive column
+compares extents instead), and deleting the source frees no shared blocks
+while a clone remains. How this compares with the other ways of saving disk
+space is in [disk space](disk_space.md).
## Shared state and caveats
@@ -141,7 +143,12 @@ deleting the source frees no shared blocks while a clone remains.
keys, was not tested.
- The host trusts each guest by UDID, so lockdown tools need the guest's
"Trust This Computer" prompt again after a new identity. Tapping Trust and
- then `idevicepair pair` worked.
+ then `idevicepair pair` worked. Copying the host's pair record
+ (`/var/db/lockdown/<old UDID>.plist`) to the new UDID might skip the
+ prompt; it needs root, so a Launchpad helper verb, and was not tried.
+- A plain clone keeps the source's ECID and MAC, so it has the same UDID and
+ competes for the same DHCP lease if both run: it is a backup, not a second
+ machine.
- A clone that later runs `restore` is personalized for its new ECID, as
usual. `cfw update-kernel` and `--update-environment` write through a host
mount and do not involve the ECID.
diff --git a/Research/Host/machine_templates.md b/Research/Host/machine_templates.md
index c3c0441e..4474cbd4 100644
--- a/Research/Host/machine_templates.md
+++ b/Research/Host/machine_templates.md
@@ -6,7 +6,9 @@ from it with `clonefile(2)` and a new identity (see
[machine identity and clone](machine_identity_and_clone.md)), so a second
machine for the same firmware costs a clone and a few hundred MB of first-boot
writes instead of a 20 GB restore. User-facing behavior is in
-[Create and run](../../Documents/Guides/create-and-run.md#templates).
+[Create and run](../../Documents/Guides/create-and-run.md#templates). What
+templates save against the alternatives that were measured, and the build and
+clone timings, are in [disk space](disk_space.md).
Code: `VPhoneKit/VPhoneCoreKit/Bundle/VPhoneMachineTemplateKey.swift` (the key,
pure), `VPhoneMachineTemplates.swift` (storage, freeze, adopt, clone, boot
@@ -496,7 +498,10 @@ conservative trim afterwards and `aggressive` were refused, and so was
booted through Launchpad to the Setup greeting with vphoned answering and 258
apps registered, no panic. With P4's snapshot deletion the 2026-10-07
prototype returned 1.19 GB for the same deletions
-(`Research/Guest/template_snapshot_deletion.md`).
+(`Research/Guest/template_snapshot_deletion.md`), and so did a full build's
+setup boot, read off the System volume (that note's "Reclaim in a template
+build"). Where these paths sit among the System volume's 16 GB, and why the
+other large items are never trimmed, is in [disk space](disk_space.md).
`vm delete` of the second of two machines cloned from a template printed
`note: template 2246f982776c (~18.92 GB) is no longer used by any machine; …`;
@@ -603,6 +608,12 @@ the profile is applied, that nothing fell short of the plan
(`VPhoneTemplateSetupOutcome.deviations(from:)`) and fails before the reboot
if anything did.
+After the reboot the verify step sometimes finds a removed app listed again
+for a few seconds (`removed apps are installed again: com.apple.tv` three
+times in one 27.0 build; News and TV for about 6 s on 26.6.2), and the next
+poll finds it gone. The step's retries absorb it; a reappearance that lasted
+past the 120 s deadline would fail the build.
+
An app left behind was seen once in two setup boots on 27.0 (2026-10-08,
Find My): "LaunchServices still lists the app after unregistration". icli's
`unregisterApp` looks at the record straight after `icli_unregister_app`, and
@@ -665,6 +676,15 @@ cleared: 13 crash report(s): duetexpertd ×11, SiriSearchFeedback ×1,
panic-full ×1` (2026-10-09 build). A report it cannot delete is a warning, as
the device name is.
+The report could be avoided at the source by stopping the DFU VM when iBoot
+prints its banner after the restore, instead of waiting for the panic. That
+would remove the panic and its report for every machine, `--no-template`
+included, and save a few seconds. It is not done: it changes the stop
+condition in both `vm create` and Launchpad (which must keep working with
+older bundles of its series), and one full creation has to confirm that
+nothing required happens in that boot; the logs show only Preboot mounted
+before launchd.
+
The reboot step's 33 s: `system.reboot` is answered at once, and the first
`ping` a moment later can still reach the old vphoned; the `processes.list`
sent after it reaches vphoned as launchd stops it and is never answered, so